The future of commerce isn't clicks — it's agents.
Users no longer browse and buy.
They delegate intent to autonomous AI agents that search, compare, negotiate, transact, optimize, and even handle post-purchase logistics — all without human intervention.
This repository is a community-driven, living collection of:
- Wild ideas & visionary concepts
- Battle-tested best practices
- Agent architectures & patterns
- Tools, frameworks & protocols
- Real-world experiments & case studies
Whether you're a builder, researcher, founder, or just someone with a crazy-good idea for agentic commerce — this is your home.
- What is Agentic Commerce?
- Core Ideas & Patterns
- Best Practices
- Tools & Frameworks
- Case Studies & Real-World Examples
- Open Challenges & Research Directions
- Contributing
- License
Agentic commerce is the shift from "human-in-the-loop" e-commerce to "agent-in-the-loop" commerce:
- Personal shopping agents that understand your taste, budget, values, and schedule
- Negotiation agents that haggle in real time across multiple marketplaces
- Supply-chain agents that autonomously reorder, reroute, and optimize inventory
- Multi-agent marketplaces where seller agents and buyer agents negotiate directly
- Autonomous checkout, fraud detection, returns, and loyalty programs
Think: “Buy me the best noise-cancelling headphones under $300 that ship in 2 days” → agent does everything.
(Share your architectural patterns, agent swarms, memory systems, tool-use strategies, etc.)
- Intent → Agent Delegation Pattern – How to turn vague user intent into executable agent workflows
- Multi-Agent Negotiation Protocols – Buyer vs Seller vs Marketplace agents
- Trust & Verification Layers – How agents prove they actually got the best deal
- Memory & Personalization Loops – Long-term user preference agents
- Add your idea here via PR!
(Real lessons from production or serious prototypes)
- Prompt engineering for commerce agents
- Tool-calling & API orchestration strategies
- Error handling & fallback mechanisms in autonomous transactions
- Privacy-preserving agent design
- Cost & latency optimization for high-volume agent runs
- Add your best practice here via PR!
(Feel free to add new entries)
- LangGraph / LangChain – Agent orchestration
- AutoGen / CrewAI / MetaGPT – Multi-agent collaboration
- Camel-AI – Role-playing agent frameworks
- OpenAI Swarm – Lightweight multi-agent orchestration
- Anthropic Computer Use + commerce APIs
- Blockchain / Payment agents (Crossmint, Stripe Agents, etc.)
- Protocol layer (e.g. Agentic Commerce Protocol proposals)
- Agent Payments Landscape – Living comparison of AP2, ACP, x402, MPP, and UCP across authorization, commerce, and settlement layers. Verified sources, protocol stack diagram.
- Agent Verifier – Spending verification for AI agents. Policy-based constraints, AP2 mandate validation, and budget enforcement. TypeScript, npm published.
- Add your favorite tool here via PR!
- OpenAI Instant Checkout + ACP in ChatGPT — Users describe what they want, the agent researches across the web, surfaces products from Etsy/Shopify merchants, and completes the full purchase inside the chat via Stripe-powered Instant Checkout. OpenAI open-sourced the Agentic Commerce Protocol (ACP) for broader adoption.
- Google Gemini "Buy for Me" / Agentic Checkout — Describe needs in plain language, and the agent searches Walmart, Wayfair, Target, Shopify, etc., checks real-time inventory/pricing/loyalty offers, and executes purchases with Google Pay. Powered by the Universal Commerce Protocol (UCP).
- Walmart + Google Gemini Full Shopping Experience — Walmart co-developed the UCP open standard with Shopify, Etsy, Target, and Wayfair. End-to-end shopping (discovery → comparison → basket → checkout) entirely inside Google Gemini.
- Perplexity "Buy with Pro" — AI search → personalized recommendations → one-click in-chat checkout directly in Perplexity. PayPal handles payments/merchant-of-record duties; free shipping on eligible orders.
- Crossmint Telegram Shopping Agents — Text a need in Telegram, the agent autonomously searches, compares options across catalogs, presents matches, and completes checkout with one tap (credit card, ACH, or stablecoins).
- Anthropic Claude as AI Travel Agent (Crossmint MCP) — Natural-language conversation to research, compare, and book flights autonomously via Crossmint payments (USDC on Solana or cards).
- Flight Deal Sniper Agents (Crossmint-powered) — Set criteria (e.g., "NYC–London under $400, max 1 stop"). The agent monitors daily price changes across platforms and auto-purchases when thresholds are met.
- Amazon Rufus AI Shopping Assistant — Conversational agent inside the Amazon app/site that researches products, answers complex queries, and drives purchases. Early reports suggest meaningfully higher conversion rates among Rufus users.
- Walmart Agentic Super Agents — Includes autonomous inventory agents using computer vision/shelf sensors that trigger restocking with zero human steps, and predictive shipping agents that forward-deploy stock before orders arrive. Early pilots report significant reductions in stockouts and faster same-day delivery.
- Sephora Digital Beauty Consultant — In-store/app agents scan skin tone/products and autonomously suggest routines + virtual try-ons with real-time personalization.
- H&M AI-Powered Retail Optimization — AI agents analyze customer interactions and sales data, delivering personalized outreach and inventory optimization. Reported improvements in basket size and conversion rates.
- eBay Mercury Platform — Real-time personalized shopping guidance and recommendation agents that act across browsing sessions.
-
OpenAI Agentic Commerce Protocol (ACP) — Open protocol enabling AI agents to discover, compare, and transact with merchants
-
Google Universal Commerce Protocol (UCP) — Open standard co-developed with Walmart, Shopify, Etsy, Target, and Wayfair for agent-driven commerce
-
x402 Payment Protocol — HTTP-native micropayment protocol using the
402 Payment Requiredstatus code, enabling agents to pay for API calls, content, and services programmatically with stablecoins — no API keys or subscriptions needed -
Machine Payments / Autonomous Agent Wallets — Infrastructure for machine-to-machine payments where AI agents hold wallets, authorize transactions, and settle payments autonomously (e.g., Crossmint agent wallets, Coinbase AgentKit, Stripe agent payments)
- How do agents handle conflicting user preferences across family members?
- Legal & liability questions when an agent buys the wrong thing
- Cross-platform identity and agent reputation systems
- Preventing agent spam / price manipulation in marketplaces
- Energy & cost efficiency at scale
- Add your open question or research idea here via PR!
This repo is only as good as the community makes it.
We love:
- New ideas (even half-baked ones)
- Best practices from real builds
- Tools, papers, demos, or GitHub repos
- Fixes, improvements, or translations
- Fork the repo
- Edit this README (or create a new Markdown file in the appropriate folder)
- Add your entry in the right section with a short description + link
- Open a Pull Request with a clear title
Example PR title: feat: add "Hierarchical Negotiation Agent Swarm" pattern
Not sure where it fits? Open an Issue or Discussion — we’ll help!
CONTRIBUTING.md has the full template and guidelines.
Made with ❤️ by the agentic commerce community
This repository is licensed under the Creative Commons Zero v1.0 Universal (CC0) — feel free to use anything here for your own projects.
Ready to shape the future of commerce?
🛠️ Build your first agent today — pick a pattern from Core Ideas, grab a tool from Frameworks, and ship it.
Then drop your experiment in a PR and join the community on Telegram! 🚀